面向遥感影像的通用模型预训练与自适应优化系统及方法
By employing a self-supervised pre-training architecture based on geographic coordinate hashing and multi-temporal mask reconstruction, combined with a dynamic hierarchical transfer mechanism and gradient inversion layer, the challenges of high-quality annotation costs and transfer learning in intelligent interpretation of remote sensing images are addressed. This architecture enables efficient remote sensing image feature extraction and cross-modal transfer, improving the application effectiveness in scenarios such as disaster early warning and resource surveys.
Patent Information
- Application Number
- CN202512004136.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-29
- Publication Date
- 2026-07-17
- Estimated Expiration
- 2045-12-29
AI Technical Summary
Existing intelligent interpretation technologies for remote sensing images face challenges such as high costs for high-quality annotation, failure to utilize the unique geographic coordinate information of remote sensing data in self-supervised pre-training, and the need to pre-set target domain data distribution parameters for transfer learning. These issues limit their application in critical scenarios such as disaster early warning and resource surveys.
A self-supervised pre-training architecture based on geographic coordinate hashing and multi-temporal mask reconstruction is adopted. Combined with a dynamic hierarchical transfer mechanism and gradient inversion layer, a cross-modal spectral invariant feature space is constructed. A lightweight cue learning framework is designed to enhance the robustness of spatiotemporal feature extraction and transfer learning efficiency of remote sensing images.
It significantly improves the feature extraction capability of remote sensing images in time-series analysis tasks, reduces the dependence on computing resources, enhances the stability of cross-modal transfer tasks and the generalization robustness of small sample scenarios, and strengthens the adaptability and engineering practicality of the model on different hardware platforms.
Smart Images
Figure CN121415186B_ABST
Abstract
Citation Information
Patent Citations
Remote sensing image road change identification method and device
CN114078237A
Unmanned aerial vehicle remote sensing image classification method, equipment and medium
CN119274096A